Methods, apparatus, equipment, media, and products for generating product recommendation lists

Abstract

This application relates to a method, apparatus, device, medium, and product for generating a product recommendation list. The method includes: acquiring an interaction data graph, with products as nodes and the number of users sharing the same interactive behavior event between any two nodes as edges; acquiring a product information graph, with products as nodes and the correlation degree corresponding to the common attributes between any two nodes as edges; determining the single-graph similarity matrix between products in each graph, and summing all single-graph similarity matrices to obtain a comprehensive similarity matrix; and determining products similar to the specified product from the comprehensive similarity matrix, based on the specified product, for extracting a product recommendation list from a product candidate library. This application can obtain a product recommendation list composed of similar products matching the specified product with lower system overhead, higher execution efficiency, lower implementation cost, and is suitable for deployment on independent websites.

Description

Technical Field

[0001] This application relates to the field of e-commerce information technology, and in particular to a method for generating a product recommendation list and the corresponding apparatus, computer equipment, computer-readable storage medium, and computer program product. Background Technology

[0002] Product recommendations are a frequently used application on e-commerce platforms, widely serving scenarios such as product search, product advertising, and product filtering. This can improve product matching efficiency, making it easier for merchants to sell their products and for consumers to have their needs met.

[0003] Traditional product recommendation methods, such as Item-CF (Item Collaboration Filter) algorithms, employ behavior-based recall strategies, recommending items similar to users' previous preferences based on user behavior data. In practice, however, data sparsity becomes a problem when user behavior data is limited. Furthermore, newly listed products may require a "cold start" due to a lack of user behavior data. In these situations, traditional product recommendation methods often fail to perform well and produce rich and accurate recommendations.

[0004] In addition, traditional product recommendation methods are often based on deep learning models, which rely on massive training samples for training with high system overhead. Therefore, their preparation efficiency is low and their preparation cost is high.

[0005] When traditional product recommendation methods are applied to e-commerce platforms based on independent websites, the aforementioned drawbacks become even more pronounced because independent websites are generally used to realize individual online stores, and product information and user behavior data are not shared between different online stores.

[0006] In view of this, the applicant has taken a different approach and explored new ideas for realizing corresponding product recommendations, and thus submitted this application. Summary of the Invention

[0007] The purpose of this application is to solve the above-mentioned problems by providing a method for generating a product recommendation list and corresponding apparatus, computer equipment, computer-readable storage medium, computer program product, etc. To meet the various purposes of this application, the following technical solutions are adopted:

[0008] In one respect, a method for generating a product recommendation list is provided to suit one of the purposes of this application, comprising the following steps:

[0009] Obtain the interaction data graph. Each interaction data graph uses products in the product candidate library as nodes, and the number of users with the same interactive behavior event between any two nodes is the edge between those two nodes.

[0010] Obtain product information graphs. Each product information graph uses products in the product candidate library as nodes and the correlation degree between the common attributes of products in any two nodes as edges.

[0011] Determine the single-image similarity matrix between products in each interactive data image and each product information image, and summarize all single-image similarity matrices to obtain the comprehensive similarity matrix between products;

[0012] Based on the specified product, identify products that are similar to the specified product from the comprehensive similarity matrix, and extract a recommended product list from the product candidate library.

[0013] Optionally, obtaining the interactive data graph includes the following steps:

[0014] Obtain user behavior data corresponding to the products in the product candidate library;

[0015] The user behavior data is cleaned to obtain interaction behavior data, which includes the accessed product, the user accessing the product, the behavior type of the interaction behavior event triggered by the user, and different behavior types correspond to different business links of the same preset business process.

[0016] Based on the interaction behavior data, an interaction data graph corresponding to each behavior type is constructed.

[0017] Optionally, based on the interaction behavior data, an interaction data graph corresponding to each behavior type is constructed, including the following steps:

[0018] Determine the target behavior type to be constructed in the interaction data graph, and filter out the corresponding interaction behavior data for that behavior type;

[0019] Create an interaction data graph corresponding to the target behavior type, and establish nodes corresponding to the products in the product candidate library within it;

[0020] Based on the selected interaction behavior data, count the number of users in the interaction behavior data corresponding to any two products, and determine the number of users in the data as the edge of the node corresponding to the two products.

[0021] Optionally, obtaining product information images includes the following steps:

[0022] Obtain multi-category product attribute data corresponding to the products in the product candidate library;

[0023] For each type of product attribute data with a unique common attribute, a corresponding product information graph is constructed. Each product information graph uses products in the product candidate library as nodes. The correlation between any two nodes is represented by a binary value based on whether the common attributes between their respective products are the same. The correlation is used as the edge between the two nodes.

[0024] For various types of product attribute data that contain multiple common attributes, a corresponding product information graph is constructed. Each product information graph takes products in the product candidate library as nodes, and the sum of the number of overlaps of common attributes between products of any two nodes and the number of overlaps of attribute values ​​corresponding to the common attributes is used as the correlation degree, and the correlation degree is used as the edge between the two nodes.

[0025] Optionally, the single-image similarity matrix between products in each interactive data graph and each product information graph is determined separately, and the comprehensive similarity matrix between products is obtained by summing all single-image similarity matrices, including the following steps:

[0026] Each interactive data graph and each product information graph is mapped to a high-dimensional information matrix space to obtain the corresponding high-dimensional information matrices.

[0027] Perform nonnegative matrix decomposition on each high-dimensional information matrix to obtain the basis matrix and coefficient matrix;

[0028] Reconstruct the basis matrix and coefficient matrix corresponding to each high-dimensional information matrix into a single graph similarity matrix;

[0029] The weighted summation of the individual image similarity matrices is used to obtain the comprehensive similarity matrix.

[0030] Optionally, based on the specified product, determine products similar to the specified product from the comprehensive similarity matrix, and extract a product recommendation list from the product candidate library, including the following steps:

[0031] Responding to a product matching request submitted by a terminal device, obtain the specified product corresponding to the request;

[0032] The row vector corresponding to the specified product is obtained by querying the comprehensive similarity matrix. Products with similarity values ​​higher than a preset threshold are selected from the row vector as target products. The similarity value of each dimension in the row vector corresponds to a product in the product candidate library.

[0033] The product information corresponding to the target product is retrieved from the product candidate library, and a product recommendation list is constructed based on the product information of the target product.

[0034] The product recommendation list is pushed to the terminal device.

[0035] Optionally, the preset business process is a commodity transaction process, which includes the following multiple business steps: browsing commodities, adding commodities to the shopping cart, creating commodity orders, and paying for commodity orders.

[0036] On the other hand, to meet one of the purposes of this application, a product recommendation list generation device is provided, including a first acquisition module, a second acquisition module, a similarity calculation module, and a product recall module, wherein: the first acquisition module is used to acquire interaction data graphs, each interaction data graph taking products in the product candidate library as nodes, and the number of users sharing the same interactive behavior event between any two nodes as the edge between the two nodes; the second acquisition module is used to acquire product information graphs, each product information graph taking products in the product candidate library as nodes, and the correlation degree corresponding to the common attributes between any two nodes as the edge; the similarity calculation module is used to determine the single graph similarity matrix between products in each interaction data graph and each product information graph, and summarize all single graph similarity matrices to obtain a comprehensive similarity matrix between products; the product recall module is used to determine products similar to the specified product from the comprehensive similarity matrix based on the specified product, and extract a product recommendation list from the product candidate library.

[0037] Optionally, the first acquisition module includes: a data acquisition submodule, used to acquire user behavior data corresponding to products in the product candidate library; a data cleaning submodule, used to clean the user behavior data to obtain interaction behavior data, the interaction behavior data including the accessed product, the user accessing the product, the behavior type of the interaction behavior event triggered by the user, and different behavior types corresponding to different business links of the same preset business process; and a construction acquisition submodule, used to construct an interaction data graph corresponding to each behavior type based on the interaction behavior data.

[0038] Optionally, the construction acquisition submodule includes: a data filtering unit, used to determine the target behavior type of the interaction data graph to be constructed, and filter out the interaction behavior data corresponding to the behavior type; a node processing unit, used to create an interaction data graph corresponding to the target behavior type, and establish nodes corresponding to the products in the product candidate library therein; and an edge processing unit, used to count the number of common users in the interaction behavior data corresponding to any two products based on the filtered interaction behavior data, and determine the number of common users as the edge of the node corresponding to the two products.

[0039] Optionally, the second acquisition module includes: an information acquisition submodule, used to acquire multi-category product attribute data corresponding to products in the product candidate library; a first construction submodule, used to construct corresponding product information graphs for each type of product attribute data with unique common attributes, wherein each product information graph uses products in the product candidate library as nodes, and the correlation between any two nodes is represented by a binary value based on whether the common attributes between their respective products are the same, and the correlation is used as the edge between the two nodes; and a second construction submodule, used to construct corresponding product information graphs for each type of product attribute data containing multiple common attributes, wherein each product information graph uses products in the product candidate library as nodes, and the correlation is the sum of the number of overlaps in the common attributes between any two nodes and the number of overlaps in the attribute values ​​corresponding to the common attributes, and the correlation is used as the edge between the two nodes.

[0040] Optionally, the similarity operation module includes: a vector transformation submodule, used to map each interactive data graph and each product information graph to a high-dimensional information matrix space to obtain the corresponding high-dimensional information matrices; a matrix decomposition submodule, used to perform non-negative matrix decomposition on each high-dimensional information matrix to obtain a basis matrix and a coefficient matrix; a matrix reconstruction submodule, used to reconstruct the basis matrix and coefficient matrix corresponding to each high-dimensional information matrix into a single-graph similarity matrix; and a weighted summation submodule, used to weighted sum the single-graph similarity matrices to obtain a comprehensive similarity matrix.

[0041] Optionally, the product recall module includes: a request response submodule, used to respond to a product matching request submitted by a terminal device and obtain the specified product corresponding to the request; a similarity matching submodule, used to query the row vector corresponding to the specified product from the comprehensive similarity matrix, and filter out products with similarity values ​​higher than a preset threshold from the row vector as target products, wherein the similarity value of each dimension in the row vector corresponds to a product in the product candidate library; a list construction submodule, used to call the product information corresponding to the target product from the product candidate library and construct a product recommendation list based on the product information of the target product; and a list push submodule, used to push the product recommendation list to the terminal device.

[0042] Optionally, the preset business process is a commodity transaction process, which includes the following multiple business steps: browsing commodities, adding commodities to the shopping cart, creating commodity orders, and paying for commodity orders.

[0043] In another aspect, a computer device provided for one of the purposes of this application includes a central processing unit and a memory, the central processing unit being used to invoke and run a computer program stored in the memory to perform the steps of the product recommendation list generation method described in this application.

[0044] In another aspect, a computer-readable storage medium is provided to suit another purpose of this application, which stores, in the form of computer-readable instructions, a computer program implemented according to the described product recommendation list generation method, which, when invoked by a computer, performs the steps included in the method.

[0045] In another aspect, a computer program product provided for another purpose of this application includes a computer program / instructions that, when executed by a processor, implement the steps of the product recommendation list generation method described in any embodiment of this application.

[0046] Compared to existing technologies, this application has several advantages, including at least the following: This application employs a graph structure. On one hand, it acquires an interaction data graph generated based on different interactive events corresponding to user behavior data. On the other hand, it acquires a product information graph generated based on different shared attributes corresponding to product information, thereby obtaining heterogeneous information related to product recommendations. Then, based on the single-graph similarity matrices obtained through independent calculations of each graph structure, a comprehensive similarity matrix is ​​synthesized. Based on the comprehensive similarity matrix, similar products in the product candidate library are recalled for a specified product, resulting in a product recommendation list. Therefore, from an information perspective, determining the similarity between pairs of products based on both product information and user behavior data ensures an accurate description of the matching degree between products. From a technical implementation perspective, the process of determining the single-graph similarity matrix for each graph and finally determining the comprehensive similarity matrix allows the entire process to efficiently and quickly obtain results based on matrix operations, resulting in low computational load and high execution efficiency. Therefore, this application can obtain a product recommendation list composed of similar products matching a specified product with lower system overhead, higher execution efficiency, lower implementation cost, and is suitable for deployment on independent websites. Attached Figure Description

[0047] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0048] Figure 1 This is a flowchart illustrating a typical embodiment of the product recommendation list generation method of this application.

[0049] Figure 2 This is a flowchart illustrating the process of obtaining a comprehensive similarity matrix based on an interactive data graph and a product information graph, as described in an embodiment of this application.

[0050] Figure 3 This is a flowchart illustrating the process of obtaining an interaction data graph based on user behavior data in an embodiment of this application.

[0051] Figure 4 This is a flowchart illustrating the process of constructing an interaction data graph corresponding to a behavior type, as described in an embodiment of this application.

[0052] Figure 5 This is a flowchart illustrating the process of constructing a product information map based on product attribute data in an embodiment of this application.

[0053] Figure 6 This is a flowchart illustrating the process of pushing a product recommendation list in response to a product matching request, as described in an embodiment of this application.

[0054] Figure 7 A schematic diagram of the device for generating a product recommendation list for this application;

[0055] Figure 8 This is a schematic diagram of the structure of a computer device used in this application. Detailed Implementation

[0056] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application.

[0057] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this application means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.

[0058] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0059] Those skilled in the art will understand that the terms "client," "terminal," and "terminal device" as used herein include both devices that receive wireless signals, devices that only possess wireless signal receiver capabilities without transmission capabilities, and devices with receiving and transmitting hardware, devices that have receiving and transmitting hardware capable of bidirectional communication over a bidirectional communication link. Such devices may include: cellular or other communication devices such as personal computers or tablets, having single-line displays, multi-line displays, or cellular or other communication devices without multi-line displays; PCS (Personal Communications Service) that can combine voice, data processing, fax, and / or data communication capabilities; PDA (Personal Digital Assistant) that may include a radio frequency receiver, pager, internet / intranet access, web browser, notepad, calendar, and / or GPS (Global Positioning System) receiver; and conventional laptops and / or handheld computers or other devices that have and / or include radio frequency receivers. As used herein, "client," "terminal," and "terminal device" can be portable, transportable, installed in a means of transportation (air, sea, and / or land), or suitable and / or configured to operate locally and / or in a distributed manner, operating in any other location on Earth and / or in space. "Client," "terminal," and "terminal device" as used herein can also be a communication terminal, an internet access terminal, or a music / video playback terminal, such as a PDA, a MID (Mobile Internet Device), and / or a mobile phone with music / video playback capabilities, or a smart TV, set-top box, etc.

[0060] The hardware referred to by the names "server," "client," and "service node" in this application is essentially an electronic device with the equivalent capabilities of a personal computer. It is a hardware device with the necessary components revealed by the von Neumann architecture, such as a central processing unit (including an arithmetic logic unit and a control unit), memory, input devices, and output devices. The computer program is stored in its memory, and the central processing unit loads the program stored in the secondary storage into the main memory to run it, execute the instructions in the program, and interact with the input and output devices to complete specific functions.

[0061] It should be noted that the concept of "server" used in this application can also be extended to the case of server clusters. Based on the network deployment principles understood by those skilled in the art, the servers should be logically divided. Physically, these servers can be independent of each other but accessible through interfaces, or they can be integrated into a single physical computer or a computer cluster. Those skilled in the art should understand this flexibility and should not use it to constrain the implementation of the network deployment method in this application.

[0062] One or more of the technical features of this application, unless explicitly specified herein, can be deployed on a server and accessed by a client remotely calling the online service interface provided by the server, or can be directly deployed and run on a client for access.

[0063] Unless otherwise specified, the neural network models referenced or potentially referenced in this application may be deployed on a remote server and invoked remotely on the client, or deployed on a client with the capability to invoke directly. In some embodiments, when running on the client, the corresponding intelligence may be acquired through transfer learning in order to reduce the requirements on the client's hardware resources and avoid excessive consumption of the client's hardware resources.

[0064] Unless otherwise specified, all data involved in this application may be stored remotely on a server or on a local terminal device, as long as it is suitable for use by the technical solution of this application.

[0065] Those skilled in the art will understand that although the various methods in this application are described based on the same concept and thus present commonality among them, they can be performed independently unless otherwise specified. Similarly, the various embodiments disclosed in this application are all based on the same inventive concept; therefore, concepts expressed in the same way, as well as concepts that are appropriately changed for convenience but are expressed differently, should be understood equivalently.

[0066] Unless otherwise expressly stated, the various embodiments disclosed in this application can be combined in a cross-cutting manner to flexibly construct new embodiments, as long as such combination does not depart from the inventive spirit of this application and can meet the needs of the prior art or solve a certain deficiency in the prior art. Those skilled in the art should be aware of such modifications.

[0067] The product recommendation list generation method of this application can be programmed into a computer program product and deployed on a client or server for execution. For example, in an exemplary application scenario of this application, it can be deployed on the server of an e-commerce platform. In this way, the method can be executed by human-computer interaction with the process of the computer program product through a graphical user interface by accessing the interface opened after the computer program product is running.

[0068] Please see Figure 1 The product recommendation list generation method of this application, in its typical embodiment, includes the following steps:

[0069] Step S1100: Obtain the interaction data graph. Each interaction data graph takes the products in the product candidate library as nodes, and the number of users with the same interactive behavior event between the products of any two nodes as the edge between the two nodes.

[0070] In an exemplary application scenario, the computer program product obtained according to this application is installed and deployed on an independent site of an e-commerce platform to provide similarity matching services for products in online stores hosted on that independent site. The online store provides a product candidate library, which can be a product database storing all products in the online store, or a product preference database composed of a subset of products collected from the product database. For example, it could be a database consisting of products in the product database that have advertising rights. In summary, the product candidate library contains multiple products, and based on each product, corresponding product information and user behavior data can be obtained.

[0071] Based on the user behavior data of the products, interaction data graphs corresponding to each behavior type in the user behavior data can be generated. Therefore, each interaction data graph corresponds to one behavior type. The interaction data graphs represent the relationship information between products and the interactive behavior events generated by the user, in the form of a graph structure.

[0072] In the interactive data graph, each product in the product candidate library is stored as a node in the interactive database. The unique characteristic information of each product can be converted into an index value and stored in its corresponding node. The edge between any two nodes is represented by the total number of users associated with the corresponding interactive behavior events in the interactive data graph between the two products corresponding to those two nodes; that is, the total number of users. Specifically, when the same user performs the same user behavior on two products corresponding to two nodes, triggering the same interactive behavior event, the edge value between these two nodes can be accumulated to 1 unit. Based on this, the edges between any two nodes in the entire product candidate library can be statistically determined.

[0073] The interaction data graph can be prepared in advance and may include one or more, typically multiple interaction data graphs, each corresponding to different interaction behavior events. Multiple interaction data graphs can provide reference information extracted from user behavior data for similarity matching of products. In one embodiment, the interaction data graph can be prepared in advance for direct use. Another embodiment, disclosed later, also provides a reference method for preparing the interaction data graph based on a product candidate library, which will not be discussed here.

[0074] In one embodiment, each interactive data graph can be represented as a feature matrix, specifically as a node feature matrix and an edge feature matrix. The product feature information of each node is represented as a row vector in the node feature matrix. This product feature information can be feature values ​​extracted from the product information of the corresponding product. For example, it can be feature values ​​obtained by applying a feature extraction model to extract deep semantic information from any one or more of the product information, such as the product title, product attribute data, product detail text, and product images. The number of users shared by each node with other nodes forms the row vector of that node in the edge feature matrix. It is easy to see that the edge feature matrix is ​​a symmetric matrix. Representing the interactive data graph as a feature matrix facilitates matrix operations, thereby improving computational efficiency.

[0075] Step S1200: Obtain product information graphs. Each product information graph uses products in the product candidate library as nodes and the correlation degree corresponding to the common attributes between any two products of each node as edges.

[0076] Based on the products in the product candidate library, their product information can be obtained, specifically the product attribute data within that information. For example, the product attribute data may include any one or more specific categories among the product's category, brand, name, and description attributes. Each specific category's product attribute data can correspond to the creation of a product information image.

[0077] In each of the aforementioned product information graphs, each product in the product candidate library is represented as a node. The unique feature information of the product can be converted into an index value and stored in its corresponding node. The edge between any two nodes is represented by the degree of association between the two products corresponding to the two nodes, which is determined by the association relationship between the product attribute data of the specific category corresponding to the product information graph.

[0078] The product attribute data for each specific category can be divided into two main categories. The first category consists of products with unique common attributes, such as category, brand, and name. Products generally have a single category, brand, and name, and therefore can be represented by a single attribute name. Thus, the product attribute data for the corresponding specific category can be classified as product attribute data with unique common attributes. The second category consists of products with non-unique common attributes, such as the descriptive attributes mentioned above. Each product typically contains multiple attribute names corresponding to multiple descriptive attributes. For example, clothing may include multiple descriptive attributes such as color, style, and size. For this type of product attribute data, it can be independently classified as product attribute data containing multiple common attributes.

[0079] In one embodiment, the correlation degree, which is suitable for a first major category with a unique common attribute, can be represented by a binary numerical value based on whether the product attribute data of any two nodes corresponding to the same specific category have the same attribute value. For example, when the specific category is product category, if the product category of the two nodes is "tops", the correlation degree between the two nodes can be represented by the value 1; if the product categories of the two nodes are different, such as one being "tops" and the other being "long johns", the correlation degree between the two nodes can be represented by the value 0. Accordingly, the edges in the product information graph corresponding to the product attribute data of a specific category are represented in the form of binary numerical values.

[0080] In another embodiment, to adapt to the second category where the shared attributes are not unique, the intersection of the same attribute names of the two products and the intersection of the same attribute values ​​of the two products can be calculated based on the product attribute data of the same specific category between any two nodes. The values ​​corresponding to the two intersections are added together as the degree of association between the two products, i.e., the two nodes, which effectively represents the edge between the two nodes.

[0081] In summary, based on the product attribute data corresponding to each product information map, the relationships between attributes of different products can be quantified into numerical values, which are used as the degree of correlation between the corresponding products and to represent the edge information between products.

[0082] The product infographic can be pre-prepared and may include one or more, typically multiple product infographics, each corresponding to different specific categories of product attribute data. Multiple product infographics can provide reference information extracted from product attribute data for product similarity matching. In one embodiment, the product infographic can be pre-prepared for direct use. Another embodiment, disclosed later, also provides a reference method for preparing the product infographic based on a product candidate library, which will not be discussed here.

[0083] In one embodiment, each product information graph can be represented as a feature matrix, specifically as a node feature matrix and an edge feature matrix. The product-specific feature information of each node is represented as a row vector in the node feature matrix. This product feature information can be feature values ​​extracted from the product information of the corresponding product. For example, it can be feature values ​​obtained by applying a feature extraction model to extract deep semantic information from any one or more of the product information, such as the product title, product attribute data, product detail text, and product images. The correlation degree between each node and other nodes forms the row vector of that node in the edge feature matrix. It is easy to see that the edge feature matrix is ​​a symmetric matrix. Representing the product information graph as a feature matrix facilitates matrix operations, thereby improving computational efficiency.

[0084] Step S1300: Determine the single-image similarity matrix between products in each interactive data image and each product information image, and summarize all single-image similarity matrices to obtain the comprehensive similarity matrix between products;

[0085] For each interactive data graph or product information graph, the similarity between products can be calculated based on the node feature matrix and edge feature matrix of each graph to obtain the corresponding single graph similarity matrix.

[0086] like Figure 2 In the illustrated embodiment, this step may include the following specific steps:

[0087] Step S1310: Map each interactive data graph and each product information graph to a high-dimensional information matrix space to obtain the corresponding high-dimensional information matrices;

[0088] When calculating the similarity of each interactive data graph or each product information graph, it can be regarded as a target graph data. Accordingly, each target graph data includes a node feature matrix and a corresponding edge feature matrix.

[0089] Furthermore, a kernel function algorithm is applied to map the node feature matrix and edge feature matrix to a high-dimensional information matrix space, thereby obtaining the high-dimensional information matrix corresponding to the target graph data. Common kernel functions include linear kernel functions, polynomial kernel functions, Gaussian kernel functions, exponential kernel functions, and Laplace kernel functions, among others, and any one of them can be used.

[0090] Therefore, by applying the kernel function algorithm to each interactive data graph and each product information graph, the high-dimensional information matrix corresponding to each graph can be obtained.

[0091] Step S1320: Perform nonnegative matrix decomposition on each high-dimensional information matrix to obtain the basis matrix and coefficient matrix;

[0092] In the high-dimensional information matrix, the existence of negative numbers is meaningless. Therefore, each high-dimensional information matrix can be decomposed into a basis matrix and a coefficient matrix using the Non-negative Matrix Factorization (NMF) algorithm.

[0093] Nonnegative matrix factorization (NMF) is a type of matrix factorization algorithm. In traditional matrix factorization techniques such as eigenvalue decomposition and SVD, the resulting matrices often have negative values, which are meaningless in practical scenarios. Based on the constraint of nonnegativity, the NMF matrix factorization algorithm was developed. For any nonnegative matrix K, it can be partitioned into the product of two nonnegative matrices W and H. W is called the basis matrix, and H is called the coefficient matrix. According to the definition of matrix multiplication, each column vector in W multiplied by the corresponding column vector in H results in a column vector in matrix K, which is essentially a linear combination.

[0094] Yes, after non-negative matrix decomposition, the high-dimensional information matrix corresponding to each graph can be obtained into two non-negative matrices, making the two decomposed matrices more meaningful in practice.

[0095] Step S1330: Reconstruct the basis matrix and coefficient matrix corresponding to each high-dimensional information matrix into a single-graph similarity matrix;

[0096] To determine the single-image similarity matrix corresponding to each target image data, the basis matrix W and coefficient matrix H corresponding to each high-dimensional information matrix can be reconstructed. This reconstruction yields a single feature matrix, which is its corresponding single-image similarity matrix. In one embodiment, the following formula can be applied to each high-dimensional information matrix K... ij The corresponding basis matrix W and coefficient matrix H are reconstructed to obtain the single-graph similarity matrix.

[0097]

[0098] This formula indicates that during the reconstruction of the base matrix and coefficient matrix, the inner product of the two is calculated, and the value of each element in the single-graph similarity matrix obtained by the inner product result can be regarded as the similarity between the product corresponding to its row and the product corresponding to its column.

[0099] It is easy to understand that determining the corresponding single-image similarity matrix for each interactive data image and product information image essentially determines the degree of similarity between products based on the information of each image's corresponding dimension.

[0100] Step S1340: Sum the weighted similarity matrices of each individual image to obtain the comprehensive similarity matrix.

[0101] To synthesize the similarity matrices of each individual image, preset weights can be applied. The individual image similarity matrices are matched with corresponding weights and then summed to obtain the final matrix as the comprehensive similarity matrix. Similarly, the value of each element in the comprehensive similarity matrix represents the overall similarity between the products in the row containing that element and the products in the column containing that element. Thus, the similarity between products in the product candidate library is determined based on the interaction data graphs and product information graphs.

[0102] Based on the above process, it can be understood that the process of determining the similarity between products in the product candidate library based on the interactive data graph and the product information graph is based on matrix operations. The computational load is low, the computational efficiency is high, and the computational results can be obtained quickly, thus reducing deployment costs.

[0103] Step S1400: Based on the specified product, determine the products that are similar to the specified product from the comprehensive similarity matrix, and extract the recommended product list from the product candidate library.

[0104] When it is necessary to match similar products for a specified product, the comprehensive similarity matrix can be used for the search.

[0105] The designated product can be determined by different objects in different embodiments. For example, in one embodiment, the designated product can be determined by the merchant user to find similar products for the merchant user, making it easier for the user to find similar sources of goods. In another embodiment, the system can determine the product as the designated product by analyzing the user-triggered interaction event that affects a certain product, so as to push similar products to the user based on the product corresponding to the interaction event, thereby playing a promotional role. And so on.

[0106] After identifying the designated product, the row vector (or column vector, hereinafter the same) of the designated product in the comprehensive similarity matrix is ​​obtained. It is easy to understand that each dimension, i.e., each element, in this row vector corresponds to each candidate product in the product candidate library, representing the similarity value between the designated product and the corresponding candidate product. Therefore, multiple candidate products with similarity values ​​higher than a preset threshold are identified, or the top N (N is a positive integer greater than 0, which can be arbitrarily preset) candidate products with high similarity values ​​are identified, and these candidate products are used to construct a product recommendation list. The product recommendation list may further include corresponding product information obtained from the product database or from the product candidate library.

[0107] The product recommendation list can be used in different ways depending on the actual application. For example, in one embodiment, the product recommendation list is used as coarse recall data and transmitted to an interface that performs fine sorting of the products in the list to obtain the final result, serving any task such as product search, product advertising, and product recommendation. In another embodiment, the product recommendation list can be directly pushed as the final result to the terminal devices accessing the online store to achieve the effect of rapid advertising.

[0108] Based on the above embodiments, it is easy to understand that, compared with the prior art, this application has many advantages, including at least:

[0109] This application employs a graph structure. On one hand, it acquires an interaction data graph generated based on different interactive events corresponding to user behavior data. On the other hand, it acquires a product information graph generated based on different shared attributes corresponding to product information. This yields heterogeneous information related to product recommendations. Then, based on the single-graph similarity matrices obtained through independent computation of each graph structure, a comprehensive similarity matrix is ​​synthesized. This comprehensive similarity matrix is ​​used to recall similar products from the product candidate library for a specified product, resulting in a product recommendation list. Therefore, from an information perspective, determining the similarity between pairs of products based on both product information and user behavior data ensures an accurate description of the matching degree between products. From a technical implementation perspective, the process of determining the single-graph similarity matrix for each individual graph and ultimately determining the comprehensive similarity matrix allows the entire process to efficiently and quickly obtain results based on matrix operations, resulting in low computational cost and high execution efficiency. Therefore, this application can obtain a product recommendation list consisting of similar products matching a specified product with lower system overhead, higher execution efficiency, and lower implementation cost, making it suitable for deployment on independent websites.

[0110] Based on any of the above embodiments, please refer to Figure 3 Step S1100, obtaining the interactive data graph, includes the following steps:

[0111] Step S1110: Obtain user behavior data corresponding to the products in the product candidate library;

[0112] The user behavior data mentioned above is mainly used to record historical data corresponding to users' interactive behaviors in online stores on e-commerce platforms. User behavior data within a preset time range can be obtained as the basic material for constructing an interaction behavior graph. The preset time range can be any duration, such as three days, seven days, or half a month from the current day, and can be preset as needed by those skilled in the art.

[0113] Step S1120: Clean the user behavior data to obtain interaction behavior data. The interaction behavior data includes the accessed product, the user accessing the product, and the behavior type of the interaction behavior event triggered by the user. Different behavior types correspond to different business links of the same preset business process.

[0114] The acquired user behavior data is generally represented in a certain data format, typically containing information corresponding to a specific business step in a pre-defined business process for a particular product, indicating the existence of a behavioral event of a corresponding type to that business step. To meet the requirements for generating interaction data graphs, the user behavior data needs to be cleaned.

[0115] In one embodiment, from each piece of user behavior data corresponding to each product, the accessed product (Product), the user who accessed the product (User), and the action type (Action) corresponding to the interactive event triggered by the user for the product are extracted. That is, the business process corresponding to the interactive event is applied to the product. Then, the product, user, and action type are constructed into mapping relationship data, for example, represented as a triple {Product; User; Action}. In this way, data cleaning of the user behavior data is achieved.

[0116] As can be seen, each interactive event corresponds to a behavior type, which represents a corresponding business operation performed on the corresponding product, and can be a corresponding business link in the corresponding preset business process.

[0117] In one embodiment, the preset business process is a product transaction process set in an e-commerce platform, which includes the following multiple business steps: browsing products, adding products to the shopping cart, creating a product order, and paying for the product order. Different interactive behavior events are divided according to the different business steps included in the product transaction process, each corresponding to a different behavior type. This allows the triple to represent a user's operation of a corresponding behavior type on a product, thereby obtaining interactive behavior data corresponding to different behavior types.

[0118] Step S1130: Based on the interaction behavior data, construct an interaction data graph corresponding to each behavior type.

[0119] After obtaining the triples of cleaned user behavior data, interaction data can be used to directly construct interaction data graphs corresponding to each behavior type based on the data contained in these triples. In one embodiment, an interaction data graph is established according to a behavior type corresponding to each business step. Taking a product transaction process as an example, four transaction data graphs can be obtained corresponding to the steps of browsing the product, adding the product to the shopping cart, creating a product order, and paying for the product order. As mentioned above, the nodes in the transaction data graph indicate specific products, and the edges between nodes are determined by the number of users with the same behavior type corresponding to the corresponding products.

[0120] As can be seen from the above embodiments, based on the user behavior data corresponding to the products in the product candidate library, after data cleaning, triples of their interaction behavior data can be obtained. Based on these triples, interaction data graphs corresponding to various behavior types can be directly constructed according to different behavior types. Each interaction data graph thus represents the edge connection relationship between products in the product candidate library determined by the number of shared users based on the behavior type corresponding to the interaction data graph. This allows the interaction data graph to reflect the association information between products in different behavior types, making it convenient to examine the similarity between products based on various different behavior types.

[0121] Based on any of the above embodiments, please refer to Figure 4 Step S1130, constructing an interaction data graph corresponding to each behavior type based on the interaction behavior data, includes the following steps:

[0122] Step S1131: Determine the target behavior type of the interaction data graph to be constructed, and filter out the corresponding interaction behavior data for that behavior type;

[0123] When it is necessary to construct an interaction data graph corresponding to a target behavior type, first filter out the interaction behavior data corresponding to the target behavior type from the interaction data obtained after data cleaning, i.e., the triples in the previous example.

[0124] Step S1132: Create an interaction data graph corresponding to the target behavior type, and establish nodes corresponding to the products in the product candidate library within it;

[0125] Based on a preset graph structure, a corresponding interaction data graph is created for the target behavior type. Then, each product in the product candidate library is stored as a node within the graph. To represent a product as a node, its unique characteristic information, such as its product ID, SKU, or SPU, can be converted into a unique index value and stored in the corresponding node. In another embodiment, the feature information obtained by extracting product information through a feature extraction model can also be associated and stored in the corresponding node for easy retrieval.

[0126] Step S1133: Based on the filtered interaction behavior data, count the number of users in the interaction behavior data corresponding to any two products, and determine the number of users in the data as the edge of the node corresponding to the two products.

[0127] To determine the edges between nodes in the interaction data graph, the interaction behavior data of all corresponding target behavior types between the two products corresponding to each pair of nodes in the interaction data graph are statistically analyzed, and the number of users in common between the two is counted. This number of users in common is used as the edge between the two nodes.

[0128] The interaction data graph corresponding to each target behavior type can be constructed according to the above process, thereby directly constructing each interaction data graph corresponding to different behavior types.

[0129] As can be seen from the above embodiments, this application can directly construct interactive data graphs corresponding to various behavior types based on the interactive behavior data obtained after data cleaning, according to a preset graph structure. The interactive data graphs can effectively represent the association information between products determined with their corresponding behavior types as reference dimensions.

[0130] Based on any of the above embodiments, please refer to Figure 5 Step S1200, obtaining the product information image, includes the following steps:

[0131] Step S1210: Obtain multi-category product attribute data corresponding to the products in the product candidate library;

[0132] Each product in the product candidate library includes multiple product information categories according to the e-commerce platform's product specifications. This product information includes a first category of product attribute data, such as product category, brand, and name, and may also include a second category of product attribute data, which provides a more comprehensive description of various product characteristics. The attributes in the first category are unique; each attribute corresponds to a specific category. Different products provide product attribute data based on common attributes corresponding to the same specific category. The second category of product attribute data directly contains multiple descriptive attributes, meaning different products may have product attribute data corresponding to multiple common attributes. This product attribute data, as part of the product information, is stored in a database associated with the online store, such as the product candidate library. It can be directly accessed when needed.

[0133] Step S1220: For each type of product attribute data with a unique common attribute, construct the corresponding product information graph. Each product information graph takes the products in the product candidate library as nodes, and the correlation between any two nodes is represented by a binary value based on whether the common attributes between the products of each node are the same. The correlation is used as the edge between the two nodes.

[0134] For the product attribute data of each specific category within the first major category, where the common attribute is unique, when constructing the corresponding product information graph for each specific category, each product information graph can still use products from the product candidate library as nodes, storing products as nodes in the manner described above. However, the edge between any two nodes can be assigned different values ​​based on whether the common attribute between the products of the two nodes is the same. Specifically, it refers to whether the attribute values ​​corresponding to the common attributes of the two products are the same. When the attribute values ​​are the same, the correlation between the two nodes can be determined to be 1; when the attribute values ​​are different, the correlation between the two nodes can be determined to be 0. This achieves the representation of the correlation between two nodes using binary numerical values, and the correlation is stored as the edge between the two nodes. Based on this principle, the edge information between nodes in the product information graph of each specific category of the first major category can be established, thereby completing the direct construction of the product information graph for each specific category.

[0135] Therefore, it is easy to understand that in this embodiment, product information graphs can be established for each specific category, such as product type, brand, and name. Then, based on whether the product type, brand, and name are the same, the corresponding edge information of each product information graph can be obtained.

[0136] Step S1230: For various types of product attribute data that contain multiple common attributes, construct the corresponding product information graph. Each product information graph takes products in the product candidate library as nodes, and uses the sum of the number of overlaps of common attributes between the products of any two nodes and the number of overlaps of the attribute values ​​corresponding to the common attributes as the correlation degree, and uses the correlation degree as the edge between the two nodes.

[0137] For the product attribute data of each category in the second major category where the common attributes are not unique, in this embodiment, the second major category may only be a single specific category, composed of data corresponding to the product's descriptive attributes. When constructing the product information graph corresponding to the second major category, the product information graph can still use products in the product candidate library as nodes, storing products as nodes in the manner described above. However, the edge between each pair of nodes can be determined by summing the number of overlaps in the common attributes between the products of the two nodes and the number of overlaps in the corresponding numerical values ​​of the common attributes. This sum is then used to determine the correlation between the two nodes, and the correlation is stored as the edge between the two nodes. Based on this principle, the edge information between nodes in the product information graph of a single specific category of the second major category can be established, thereby completing the direct construction of the product information graph of each specific category.

[0138] Based on the above embodiments, it can be seen that this application can directly construct multiple product information graphs corresponding to specific categories based on the product attribute data obtained after data cleaning, according to the preset graph structure. Different specific categories are used to correspond to different types of product attribute data. The product information graphs constructed in this way can effectively represent the association information between products based on different product attributes.

[0139] Based on any of the above embodiments, please refer to Figure 6 Step S1400, which involves determining products similar to the specified product from the comprehensive similarity matrix and extracting a recommended product list from the product candidate library, includes the following steps:

[0140] Step S1410: Respond to the product matching request submitted by the terminal device and obtain the specified product corresponding to the request;

[0141] When a consumer user accesses a product on their device in an online store, the page they access can trigger a product matching request via a backend command. This request includes the currently accessed product as the specified product and is sent to the server of the independent website. The server receives the product matching request, parses it, and obtains the specified product.

[0142] Step S1420: Query the row vector corresponding to the specified product from the comprehensive similarity matrix, and select products with similarity values ​​higher than a preset threshold from the row vector as target products. The similarity value of each dimension in the row vector corresponds to a product in the product candidate library.

[0143] As mentioned earlier, a comprehensive similarity matrix corresponding to the product candidate library has been predetermined. This matrix represents the similarity values ​​between any two products in the product candidate library, and each row vector represents the similarity between the product corresponding to its row and each corresponding product in the product candidate library. Therefore, by extracting the row vector corresponding to the specified product from this comprehensive similarity matrix, the similarity value stored in each dimension (element) of this row vector represents the similarity between the specified product and a corresponding product in the product candidate library. Thus, products with similarity values ​​higher than a preset threshold can be selected from the row vectors as target products for constructing the product recommendation list. The preset threshold can be a measured threshold or an empirical threshold, which can be set as needed by those skilled in the art.

[0144] Step S1430: Retrieve the product information corresponding to the target product from the product candidate library, and construct a product recommendation list based on the product information of the target product;

[0145] The product candidate library can pre-store product information for each product, including product title, product image, product price, product page link, and other specific information. Accordingly, after determining the target product, the corresponding product information for each target product is further obtained from the product candidate library. This product information is then constructed into a structure according to a preset format specification, and the structures of each target product are encapsulated in a product recommendation list.

[0146] Step S1440: Push the product recommendation list to the terminal device.

[0147] Finally, the product recommendation list is pushed to the terminal device in response to the product matching request. The terminal device then parses and displays the product title, product image, and product price of each target product in the product recommendation list, and associates the product image and product title with the product page link, so that users can easily enter the corresponding target product page by touching the product image or product title.

[0148] As can be seen from the above embodiments, based on the comprehensive similarity matrix obtained in advance by this application, a product matching request can be responded to. Based on the specified product in the product matching request, similar target products that match the specified product can be quickly and efficiently determined. The product matching request can be silently sent by the background command of the user's terminal device without user intervention, thereby playing the role of delivering product advertisements to the user.

[0149] Please see Figure 7 To meet one of the purposes of this application, a product recommendation list generation device is provided, which is a functional embodiment of the product recommendation list generation method of this application. The device includes a first acquisition module 1100, a second acquisition module 1200, a similarity calculation module 1300, and a product recall module 1400, wherein: the first acquisition module 1100 is used to acquire interaction data graphs, where each interaction data graph uses products in the product candidate library as nodes, and the number of users sharing the same interactive behavior event between any two nodes is the edge between those two nodes; the second acquisition module 1200 is used to... The system acquires product information images, with each product information image having products in the product candidate library as nodes and edges corresponding to the common attributes between products of any two nodes. The similarity calculation module 1300 is used to determine the single-image similarity matrix between products in each interactive data image and each product information image, and to summarize all single-image similarity matrices to obtain a comprehensive similarity matrix between products. The product recall module 1400 is used to determine products similar to the specified product from the comprehensive similarity matrix based on the specified product, and to extract a product recommendation list from the product candidate library.

[0150] Optionally, the first acquisition module 1100 includes: a data acquisition submodule, used to acquire user behavior data corresponding to products in the product candidate library; a data cleaning submodule, used to clean the user behavior data to obtain interaction behavior data, the interaction behavior data including the accessed product, the user accessing the product, the behavior type of the interaction behavior event triggered by the user, and different behavior types corresponding to different business links of the same preset business process; and a construction acquisition submodule, used to construct an interaction data graph corresponding to each behavior type based on the interaction behavior data.

[0151] Optionally, the construction acquisition submodule includes: a data filtering unit, used to determine the target behavior type of the interaction data graph to be constructed, and filter out the interaction behavior data corresponding to the behavior type; a node processing unit, used to create an interaction data graph corresponding to the target behavior type, and establish nodes corresponding to the products in the product candidate library therein; and an edge processing unit, used to count the number of common users in the interaction behavior data corresponding to any two products based on the filtered interaction behavior data, and determine the number of common users as the edge of the node corresponding to the two products.

[0152] Optionally, the second acquisition module 1200 includes: an information acquisition submodule, used to acquire multi-category product attribute data corresponding to products in the product candidate library; a first construction submodule, used to construct corresponding product information graphs for each type of product attribute data with unique common attributes, wherein each product information graph uses products in the product candidate library as nodes, and the correlation between any two nodes is represented by a binary value based on whether the common attributes between their respective products are the same, and the correlation is used as the edge between the two nodes; and a second construction submodule, used to construct corresponding product information graphs for each type of product attribute data containing multiple common attributes, wherein each product information graph uses products in the product candidate library as nodes, and the correlation is the sum of the number of overlaps of the common attributes between any two nodes and the number of overlaps of the attribute values ​​corresponding to the common attributes, and the correlation is used as the edge between the two nodes.

[0153] Optionally, the similarity operation module 1300 includes: a vector transformation submodule, used to map each interactive data graph and each product information graph to a high-dimensional information matrix space to obtain the corresponding high-dimensional information matrices; a matrix decomposition submodule, used to perform non-negative matrix decomposition on each high-dimensional information matrix to obtain a basis matrix and a coefficient matrix; a matrix reconstruction submodule, used to reconstruct the basis matrix and coefficient matrix corresponding to each high-dimensional information matrix into a single graph similarity matrix; and a weighted summation submodule, used to weighted sum the single graph similarity matrices to obtain a comprehensive similarity matrix.

[0154] Optionally, the product recall module 1400 includes: a request response submodule, used to respond to a product matching request submitted by a terminal device and obtain a specified product corresponding to the request; a similarity matching submodule, used to query the row vector corresponding to the specified product from the comprehensive similarity matrix, and filter out products with similarity values ​​higher than a preset threshold from the row vector as target products, wherein the similarity value of each dimension in the row vector corresponds to a product in the product candidate library; a list construction submodule, used to call the product information corresponding to the target product from the product candidate library and construct a product recommendation list based on the product information of the target product; and a list push submodule, used to push the product recommendation list to the terminal device.

[0155] Optionally, the preset business process is a commodity transaction process, which includes the following multiple business steps: browsing commodities, adding commodities to the shopping cart, creating commodity orders, and paying for commodity orders.

[0156] To address the aforementioned technical problems, embodiments of this application also provide computer equipment. For example... Figure 8 The diagram shows the internal structure of a computer device. The computer device includes a processor, a computer-readable storage medium, a memory, and a network interface connected via a system bus. The computer-readable storage medium stores an operating system, a database, and computer-readable instructions. The database may store a sequence of control information. When executed by the processor, the computer-readable instructions enable the processor to implement a product search category identification method. The processor provides computing and control capabilities to support the operation of the entire computer device. The memory stores computer-readable instructions, which, when executed by the processor, enable the processor to execute the product recommendation list generation method of this application. The network interface of the computer device is used for communication with a terminal. Those skilled in the art will understand that… Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0157] In this embodiment, the processor is used to execute... Figure 7The specific functions of each module and its sub-modules are defined within the device. The memory stores the program code and various data required to execute these modules or sub-modules. The network interface is used for data transmission between the user terminal and the server. In this embodiment, the memory stores the program code and data required to execute all modules / sub-modules in the product recommendation list generation device of this application. The server can call the server's program code and data to execute the functions of all sub-modules.

[0158] This application also provides a storage medium storing computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the steps of the product recommendation list generation method of any embodiment of this application.

[0159] This application also provides a computer program product, including a computer program / instructions that, when executed by one or more processors, implement the steps of the method described in any embodiment of this application.

[0160] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. This computer program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. The aforementioned storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0161] In summary, from an information perspective, determining the similarity between pairs of products based on both product information and user behavior data ensures an accurate description of the matching degree between products. From a technical implementation perspective, the process of determining the single-image similarity matrix for each individual image and ultimately determining the comprehensive similarity matrix allows the entire process to efficiently and quickly obtain results based on matrix operations, resulting in low computational cost and high execution efficiency. Therefore, this application can obtain a product recommendation list consisting of similar products matching a specified product with lower system overhead, resulting in higher execution efficiency, lower implementation cost, and suitability for deployment on independent websites.

[0162] Those skilled in the art will understand that the steps, measures, and solutions in the various operations, methods, and processes discussed in this application can be alternated, modified, combined, or deleted. Furthermore, other steps, measures, and solutions in the various operations, methods, and processes discussed in this application can also be alternated, modified, rearranged, decomposed, combined, or deleted. Furthermore, steps, measures, and solutions in the prior art that are similar to those disclosed in this application can also be alternated, modified, rearranged, decomposed, combined, or deleted.

[0163] The above description is only a partial embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for generating a product recommendation list, characterized in that, Includes the following steps: Obtain the interaction data graph. Each interaction data graph uses products in the product candidate library as nodes, and the number of users with the same interactive behavior event between any two nodes is the edge between those two nodes. Obtain product information graphs. Each product information graph uses products in the product candidate library as nodes and the correlation degree between the common attributes of products in any two nodes as edges. The process involves determining the single-image similarity matrix between products in each interactive data graph and each product information graph, and then summing all single-image similarity matrices to obtain a comprehensive similarity matrix between products. This includes: applying a kernel function algorithm to map each interactive data graph and each product information graph to a high-dimensional information matrix space to obtain the corresponding high-dimensional information matrices; performing non-negative matrix decomposition on each high-dimensional information matrix to obtain a basis matrix and a coefficient matrix; reconstructing the basis matrix and coefficient matrix corresponding to each high-dimensional information matrix into a single-image similarity matrix, including: determining the element values ​​of the single-image similarity matrix based on the inner product of the basis matrix and the coefficient matrix; and weighted summing of the single-image similarity matrices to obtain a comprehensive similarity matrix. Based on the specified product, identify products that are similar to the specified product from the comprehensive similarity matrix, and extract a recommended product list from the product candidate library.

2. The method for generating a product recommendation list according to claim 1, characterized in that, Obtaining the interactive data graph includes the following steps: Obtain user behavior data corresponding to the products in the product candidate library; The user behavior data is cleaned to obtain interaction behavior data, which includes the accessed product, the user accessing the product, the behavior type of the interaction behavior event triggered by the user, and different behavior types correspond to different business links of the same preset business process. Based on the interaction behavior data, an interaction data graph corresponding to each behavior type is constructed.

3. The method for generating a product recommendation list according to claim 2, characterized in that, Based on the interaction behavior data, an interaction data graph corresponding to each behavior type is constructed, including the following steps: Determine the target behavior type to be constructed in the interaction data graph, and filter out the corresponding interaction behavior data for that behavior type; Create an interaction data graph corresponding to the target behavior type, and establish nodes corresponding to the products in the product candidate library within it; Based on the selected interaction behavior data, count the number of users in the interaction behavior data corresponding to any two products, and determine the number of users in the data as the edge of the node corresponding to the two products.

4. The method for generating a product recommendation list according to claim 1, characterized in that, Obtaining product information images involves the following steps: Obtain multi-category product attribute data corresponding to the products in the product candidate library; For each type of product attribute data with a unique common attribute, a corresponding product information graph is constructed. Each product information graph uses products in the product candidate library as nodes. The correlation between any two nodes is represented by a binary value based on whether the common attributes between their respective products are the same. The correlation is used as the edge between the two nodes. For various types of product attribute data that contain multiple common attributes, a corresponding product information graph is constructed. Each product information graph takes products in the product candidate library as nodes, and the sum of the number of overlaps of common attributes between products of any two nodes and the number of overlaps of attribute values ​​corresponding to the common attributes is used as the correlation degree, and the correlation degree is used as the edge between the two nodes.

5. The method for generating a product recommendation list according to claim 1, characterized in that, Based on a specified product, identify products similar to the specified product from the comprehensive similarity matrix, and extract a recommended product list from the product candidate library, including the following steps: Responding to a product matching request submitted by a terminal device, obtain the specified product corresponding to the request; The row vector corresponding to the specified product is obtained by querying the comprehensive similarity matrix. Products with similarity values ​​higher than a preset threshold are selected from the row vector as target products. The similarity value of each dimension in the row vector corresponds to a product in the product candidate library. The product information corresponding to the target product is retrieved from the product candidate library, and a product recommendation list is constructed based on the product information of the target product. The product recommendation list is pushed to the terminal device.

6. The method for generating a product recommendation list according to claim 2, characterized in that, The preset business process is a commodity transaction process, which includes the following multiple business steps: browsing commodities, adding commodities to the shopping cart, creating commodity orders, and paying for commodity orders.

7. A product recommendation list generation device, characterized in that, include: The first acquisition module is used to acquire interactive data graphs. Each interactive data graph takes a product in the product candidate library as a node and the number of users with the same interactive behavior event between any two products of each node as the edge between the two nodes. The second acquisition module is used to acquire product information graphs. Each product information graph uses products in the product candidate library as nodes and the correlation degree corresponding to the common attributes between the products of any two nodes as edges. The similarity operation module is used to determine the single-image similarity matrix between products in each interactive data graph and each product information graph, and to summarize all single-image similarity matrices to obtain a comprehensive similarity matrix between products. This includes: a vector transformation submodule, which applies a kernel function algorithm to map each interactive data graph and each product information graph to a high-dimensional information matrix space to obtain the corresponding high-dimensional information matrices; a matrix decomposition submodule, which performs non-negative matrix decomposition on each high-dimensional information matrix to obtain a basis matrix and a coefficient matrix; a matrix reconstruction submodule, which reconstructs the basis matrix and coefficient matrix corresponding to each high-dimensional information matrix into a single-image similarity matrix, including: determining the element values ​​of the single-image similarity matrix based on the inner product of the basis matrix and the coefficient matrix; and a weighted summation submodule, which weights and sums the single-image similarity matrices to obtain a comprehensive similarity matrix. The product recall module is used to identify products that are similar to a specified product from the comprehensive similarity matrix and extract a recommended product list from the product candidate library.

8. A computer device comprising a central processing unit and a memory, characterized in that, The central processing unit is used to invoke and run a computer program stored in the memory to perform the steps of the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, It stores, in the form of computer-readable instructions, a computer program implemented according to any one of claims 1 to 6, which, when invoked by a computer, executes the steps included in the corresponding method.

Images